A Machine Learning-Based Prediction Method for Dehydrogenation Performance of Borohydrides

By combining data preprocessing and feature selection with multiple machine learning models, the problems of time-consuming and costly prediction of borohydride dehydrogenation performance are solved, enabling rapid and accurate evaluation of borohydride dehydrogenation performance and supporting the research and application of borohydrides.

CN118280489BActive Publication Date: 2026-11-17GUILIN UNIV OF ELECTRONIC TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410463060.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-17
Publication Date
2026-11-17
Estimated Expiration
2044-04-17

AI Technical Summary

Technical Problem

Existing technologies for predicting the dehydrogenation performance of borohydrides are time-consuming, costly, and difficult to respond quickly to the needs of new material development. Furthermore, the selection of machine learning models and the acquisition of data are difficult, resulting in inconsistent prediction accuracy.

Method used

Using data preprocessing techniques and feature selection methods, combined with linear regression, support vector machine, decision tree and random forest models, the dehydrogenation performance of borohydrides is predicted by hydrogen bond length. Relevant features are extracted using the Material Project library and other databases for training and evaluation.

Benefits of technology

This method enables rapid and low-cost prediction of the dehydrogenation performance of borohydrides, improving research efficiency and accuracy and providing a new technical approach for the research and application of borohydrides.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118280489B_ABST
    Figure CN118280489B_ABST
Patent Text Reader

Abstract

This invention discloses a machine learning-based method for predicting the dehydrogenation performance of borohydrides, characterized by the following steps: S1 data acquisition; S2 data preprocessing; S3 feature selection and enhancement; S4 model training and evaluation; S5 dehydrogenation performance prediction. This method, employing data preprocessing and feature selection techniques, can effectively predict the dehydrogenation performance of borohydrides, providing a new technical approach for the research and application of borohydrides.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to a technology for predicting the dehydrogenation performance of borohydrides, and in particular to a method for predicting the dehydrogenation performance of hydrogen storage materials by predicting the hydrogen bond length of borohydrides. Specifically, it is a machine learning-based method for predicting the dehydrogenation performance of borohydrides. Background Technology

[0002] Borohydrides, as highly efficient hydrogen storage materials, have attracted widespread attention due to their high theoretical hydrogen storage capacity and relatively stable chemical properties. In the field of energy conversion and storage, the dehydrogenation performance of borohydrides is a key factor determining their application potential. The quality of borohydride dehydrogenation directly affects the hydrogen release rate and efficiency, which is crucial for the practical application and economic viability of hydrogen energy.

[0003] Traditional methods for evaluating the dehydrogenation performance of borohydrides rely on complex experimental tests and theoretical calculations. These methods are often time-consuming and costly, and cannot quickly respond to the development needs of new materials. With the development of machine learning technology, machine learning algorithms are being used to predict the dehydrogenation performance of borohydrides, aiming to obtain a more efficient, faster, and more accurate prediction method.

[0004] However, there are several technical challenges in using machine learning to predict the dehydrogenation performance of borohydrides. First, the structural data of borohydrides is complex, containing various borohydride bond types and lattice structures. Extracting features related to dehydrogenation performance from this complex data is a challenge. Second, to improve prediction accuracy, a large amount of labeled data is needed to train the machine learning model, but such data is often difficult to obtain. Finally, different machine learning models yield different prediction accuracies, making the selection of the appropriate model another important consideration. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a machine learning-based method for predicting the dehydrogenation performance of borohydrides. This method employs data preprocessing techniques and feature selection methods to effectively predict the dehydrogenation performance of borohydrides, providing a new technical approach for the research and application of borohydrides.

[0006] The technical solution to achieve the objective of this invention is:

[0007] A machine learning-based method for predicting the dehydrogenation performance of borohydrides includes the following steps:

[0008] S1 Data Acquisition: Acquire the raw data needed to predict hydrogen bond lengths: Specifically, extract the atomic coordinates of all hydrogen-containing materials from the Material Project library and then calculate the hydrogen bond lengths of the materials using the following formula:

[0009]

[0010] In the formula (x H y H , z H (x) represents the atomic coordinates of a hydrogen atom in the material. * y * , z * The coordinates of other atoms bonded to hydrogen atoms in the material are denoted as , and other data features, namely chemical formula, space group, electronegativity, enthalpy of formation, atomic radius, atomic coordinates, theoretical hydrogen storage capacity, atomic volume, and atomic density, are extracted from various existing databases on the Internet, such as Materials Project, JARVIS-DFT, and the high-throughput computing platform for electrochemical energy storage materials.

[0011] S2 Data Preprocessing: This involves preliminary processing of the raw data, specifically cleaning the data collected in step S1, removing incomplete or erroneous records, and performing standardization.

[0012] S2.1: Clean the data, that is, remove incomplete or incorrect records, unify the units of measurement and eliminate the influence of numerical range on the data extracted in step S1, specifically convert the length units of all features to angstroms, and retain a maximum of 5 decimal places for all data.

[0013] S2.2: Standardize the data extracted in step S1, that is, convert the chemical formula symbols of the materials into data symbols that can be recognized by computers;

[0014] S3 Feature Selection and Enhancement: Feature selection and enhancement involves selecting features that are highly correlated with hydrogen bond length. The correlation between features is analyzed using the shap function image, and nine features that contribute most to the prediction of hydrogen bond length are selected: atomic volume, atomic density, number of hydrogen atoms in the material, space set, electronegativity, heat of formation of each atom, atomic radius, difference between the sum of the radii of the two bonding atoms and the hydrogen bond length, and bonding element.

[0015] S4 Model Training and Evaluation: Four models were used for training: linear regression, support vector machine, decision tree, and random forest. The model was then evaluated based on the root mean square error (MSE) and coefficient of determination (R²). 2 The metrics quantify the predictive performance of each model, including:

[0016] S4.1: The nine features most strongly correlated with hydrogen bond length selected in step S3, namely atomic volume, atomic density, number of hydrogen atoms in the material, spatial set, electronegativity, heat of formation of each atom, atomic radius, difference between the sum of the radii of the two bonding atoms and the hydrogen bond length, and bonding elements, are used as the dataset. Then, the dataset is divided into training and testing sets in an 8:2 ratio and fed into four models: linear regression model, support vector machine model, decision tree model, and random forest model, respectively, for model training and testing.

[0017] S4.2: The root mean square error (MSE) is used to evaluate the accuracy of the model. The smaller the MSE, the higher the model accuracy. The calculation formula is as follows:

[0018]

[0019] In the formula, L pre To predict hydrogen bond length, L true Where n is the actual hydrogen bond length and n is the number of samples;

[0020] S4.3 Based on the coefficient of determination R 2 The accuracy of the evaluation model is measured by the coefficient of determination R. 2 The maximum value is 1, and the minimum value is 0. A value closer to 1 indicates a better model, while a value closer to 0 indicates a worse model. The coefficient of determination R is then calculated. 2 The regression sum of squares (SSR), residual sum of squares (SSE), and total sum of squared deviations (SST) are calculated as follows:

[0021] First, calculate the sum of squared regressions (SSR). The formula is as follows:

[0022]

[0023] In the formula, L pre To predict hydrogen bond length, This represents the average actual hydrogen bond length;

[0024] Next, calculate the sum of squared residuals (SSE). The formula is as follows:

[0025] SSE=∑(L ture -L pre ) 2 ,

[0026] In the formula, L pre To predict hydrogen bond length, L true This represents the actual hydrogen bond length.

[0027] Then calculate the total sum of squared deviations (SST) using the following formula:

[0028] SST = SSR + SSE

[0029] In the formula, SSR is the regression sum of squares, and SSE is the residual sum of squares.

[0030] Finally, the coefficient of determination R is calculated. 2 The calculation formula is:

[0031]

[0032] In the formula, SSE is the sum of squared residuals, and SST is the sum of squared total deviations.

[0033] S5 Dehydrogenation Performance Prediction: A trained model is used to predict the hydrogen bond length of new hydrogen storage material samples. Based on the hydrogen bond length prediction results, the dehydrogenation performance of the materials is evaluated, and materials with excellent dehydrogenation performance are identified.

[0034] This technical solution evaluates and optimizes the dehydrogenation efficiency of borohydrides by effectively predicting hydrogen bond length, and has the following advantages:

[0035] 1. High efficiency: Machine learning can quickly process and analyze large amounts of data, reducing the time required for traditional experimental methods and theoretical calculations, and improving research efficiency;

[0036] 2. Low cost: Compared with experimental methods, machine learning models reduce the use of experimental materials and equipment, thus lowering research costs.

[0037] This method employs data preprocessing techniques and feature selection methods, which can effectively predict the dehydrogenation performance of borohydrides, providing a new technical approach for the research and application of borohydrides. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating the method used in the embodiment.

[0039] Figure 2 The image shows the shap function graph obtained in the random forest model during feature selection in the embodiment.

[0040] Figure 3 The image shows the evaluation graphs of the four models obtained from model training and evaluation in the embodiment. Detailed Implementation

[0041] The present invention will be further described below with reference to the accompanying drawings and embodiments, but this is not intended to limit the scope of the invention.

[0042] Example:

[0043] For reference Figure 1 A machine learning-based method for predicting the dehydrogenation performance of borohydrides includes the following steps:

[0044] S1 Data Acquisition: Acquire the raw data needed to predict hydrogen bond lengths: Specifically, extract the atomic coordinates of all hydrogen-containing materials from the Material Project library and then calculate the hydrogen bond lengths of the materials using the following formula:

[0045]

[0046] In the formula (x H y H , z H (x) represents the atomic coordinates of a hydrogen atom in the material. * y * , z * The coordinates of other atoms bonded to hydrogen atoms in the material are shown. Other data features, such as chemical formula, space group, electronegativity, enthalpy of formation, atomic radius, atomic coordinates, theoretical hydrogen storage capacity, atomic volume, and atomic density, are extracted from various databases available on the Internet, including Materials Project, JARVIS-DFT, and high-throughput computing platforms for electrochemical energy storage materials. In this example, we not only focus on the single hydrogen bond length feature, but also extract multi-dimensional features, including interatomic distance, lattice parameters, and electronic structure, by comprehensively analyzing the microstructure data of hydrogen storage materials, providing more comprehensive data support for the prediction of dehydrogenation performance.

[0047] S2 Data Preprocessing: This involves preliminary processing of the raw data, specifically cleaning the data collected in step S1, removing incomplete or erroneous records, and performing standardization.

[0048] S2.1: Clean the data, that is, remove incomplete or incorrect records, unify the units of measurement and eliminate the influence of numerical range on the data extracted in step S1, specifically convert the length units of all features to angstroms, and retain a maximum of 5 decimal places for all data.

[0049] S2.2: Standardize the data extracted in step S1, that is, convert the chemical formula symbols of the materials into data symbols that can be recognized by computers;

[0050] S3 Feature Selection and Enhancement: Feature selection and enhancement involves choosing features highly correlated with hydrogen bond length. The correlation between features is analyzed using the shap function image, and nine features are selected that contribute most to predicting hydrogen bond length: atomic volume, atomic density, number of hydrogen atoms in the material, space set, electronegativity, heat of formation per atom, atomic radius, the difference between the sum of the radii of the two bonding atoms and the hydrogen bond length, and the bonding element. This example uses a random forest model. The shap function image representing the correlation between other atomic features and hydrogen bond length is shown below. Figure 2 As shown in this example, by applying the shap function for analysis, the expressive power of the feature data and the accuracy of the prediction model are enhanced.

[0051] S4 Model Training and Evaluation: Four models were used for training: linear regression, support vector machine, decision tree, and random forest. The model was then evaluated based on the root mean square error (MSE) and coefficient of determination (R²). 2 The metrics quantify the predictive performance of each model, including:

[0052] S4.1: The nine features most strongly correlated with hydrogen bond length selected in step S3, namely atomic volume, atomic density, number of hydrogen atoms in the material, spatial set, electronegativity, heat of formation of each atom, atomic radius, difference between the sum of the radii of the two bonding atoms and the hydrogen bond length, and bonding elements, are used as the dataset. Then, the dataset is divided into training and testing sets in an 8:2 ratio and fed into four models: linear regression model, support vector machine model, decision tree model, and random forest model, respectively, for model training and testing.

[0053] S4.2: The root mean square error (MSE) is used to evaluate the accuracy of the model. The smaller the MSE, the higher the model accuracy. The calculation formula is as follows:

[0054]

[0055] In the formula, L pre To predict hydrogen bond length, L true Where n is the actual hydrogen bond length and n is the number of samples;

[0056] S4.3 Based on the coefficient of determination R 2 The accuracy of the evaluation model is measured by the coefficient of determination R. 2 The maximum value is 1, and the minimum value is 0. A value closer to 1 indicates a better model, while a value closer to 0 indicates a worse model. The coefficient of determination R is then calculated. 2 The regression sum of squares (SSR), residual sum of squares (SSE), and total sum of squared deviations (SST) are calculated as follows:

[0057] First, calculate the sum of squared regressions (SSR). The formula is as follows:

[0058]

[0059] In the formula, L pre To predict hydrogen bond length, This represents the average actual hydrogen bond length;

[0060] Next, calculate the sum of squared residuals (SSE). The formula is as follows:

[0061] SSE=∑(L ture -L pre ) 2 ,

[0062] In the formula, Lpre To predict hydrogen bond length, L true This represents the actual hydrogen bond length.

[0063] Then calculate the total sum of squared deviations (SST) using the following formula:

[0064] SST = SSR + SSE

[0065] In the formula, SSR is the regression sum of squares, and SSE is the residual sum of squares.

[0066] Finally, the coefficient of determination R is calculated. 2 The calculation formula is:

[0067]

[0068] In the formula, SSE is the sum of squared residuals, and SST is the sum of squared total deviations. The model evaluation graphs for the four models obtained in this example—linear regression, support vector machine, decision tree, and random forest—are shown below. Figure 3 As shown, multiple performance evaluation metrics are used to comprehensively evaluate and optimize the model, ensuring that the model has good generalization ability and prediction accuracy, and improving the reliability and stability of the model in practical applications.

[0069] S5 Dehydrogenation Performance Prediction: A trained model is used to predict the hydrogen bond length of new hydrogen storage material samples. Based on the prediction results, the dehydrogenation performance of the materials is evaluated, and materials with excellent dehydrogenation performance are identified, thereby guiding experimental research and industrial production.

Claims

1. A method for predicting the dehydrogenation performance of borohydrides based on machine learning, characterized in that, Includes the following steps: S1 Data Acquisition: Acquire the raw data needed to predict hydrogen bond lengths: Specifically, extract the atomic coordinates of all hydrogen-containing materials from the Material Project library and then calculate the hydrogen bond lengths of the materials using the following formula: (x H , y H , z H ) are atomic coordinates of hydrogen atoms in the material, (x * , y * , z * ) are coordinates of other atoms in the material which form bonds with hydrogen atoms, other data features, i.e. chemical formula, space group, electronegativity, formation enthalpy, atomic radius, atomic coordinates, theoretical hydrogen storage capacity, atomic volume, atomic density, are extracted from various databases available on the web from Materials Project, JARVIS-DFT and High-Throughput Computational Platform for Electrochemical Energy Storage Materials; S2 Data Preprocessing: This involves preliminary processing of the raw data, specifically cleaning the data collected in step S1, removing incomplete or erroneous records, and performing standardization. S2.1: Clean the data, that is, remove incomplete or incorrect records, unify the units of measurement and eliminate the influence of numerical range on the data extracted in step S1, specifically convert the length units of all features to angstroms, and retain a maximum of 5 decimal places for all data. S2.2: Standardize the data extracted in step S1, that is, convert the chemical formula symbols of the materials into data symbols that can be recognized by computers; S3 Feature Selection and Enhancement: Feature selection and enhancement involves selecting features that are highly correlated with hydrogen bond length. The correlation between features is analyzed using the shap function image, and nine features that contribute most to the prediction of hydrogen bond length are selected: atomic volume, atomic density, number of hydrogen atoms in the material, space set, electronegativity, heat of formation of each atom, atomic radius, difference between the sum of the radii of the two bonding atoms and the hydrogen bond length, and bonding element. S4 Model Training and Evaluation: Four models were used for training: linear regression, support vector machine, decision tree, and random forest. The model was then evaluated based on the root mean square error (MSE) and coefficient of determination (R²). 2 The metrics quantify the predictive performance of each model, including: S4.1: The nine features most strongly correlated with hydrogen bond length selected in step S3, namely atomic volume, atomic density, number of hydrogen atoms in the material, spatial set, electronegativity, heat of formation of each atom, atomic radius, difference between the sum of the radii of the two bonding atoms and the hydrogen bond length, and bonding elements, are used as the dataset. Then, the dataset is divided into training and testing sets in an 8:2 ratio and input into four models: linear regression model, support vector machine model, decision tree model, and random forest model, respectively, for model training and testing. S4.2: The root mean square error (MSE) is used to evaluate the accuracy of the model. The smaller the MSE, the higher the model accuracy. The calculation formula is as follows: In the formula, L pre To predict hydrogen bond length, L true Where n is the actual hydrogen bond length and n is the number of samples; S4.3 Based on the coefficient of determination R 2 The accuracy of the evaluation model is measured by the coefficient of determination R. 2 The maximum value is 1, and the minimum value is 0. A value closer to 1 indicates a better model, while a value closer to 0 indicates a worse model. The coefficient of determination R is then calculated. 2 The regression sum of squares (SSR), residual sum of squares (SSE), and total sum of squared deviations (SST) are calculated as follows: First, calculate the sum of squared regressions (SSR). The formula is as follows: In the formula, L pre To predict hydrogen bond length, This represents the average actual hydrogen bond length; Next, calculate the sum of squared residuals (SSE). The formula is as follows: SSE=∑(L ture -L pre ) 2 , In the formula, L pre To predict hydrogen bond length, L true This represents the actual hydrogen bond length. Then calculate the total sum of squared deviations (SST) using the following formula: SST = SSR + SSE In the formula, SSR is the regression sum of squares, and SSE is the residual sum of squares. Finally, the coefficient of determination R is calculated. 2 The calculation formula is: In the formula, SSE is the sum of squared residuals, and SST is the sum of squared total deviations. S5 Dehydrogenation Performance Prediction: A trained model is used to predict the hydrogen bond length of new hydrogen storage material samples. Based on the hydrogen bond length prediction results, the dehydrogenation performance of the materials is evaluated, and materials with excellent dehydrogenation performance are identified.